JP2005235307A - Logic-in-memory circuit using magneto-resistance effect element - Google Patents

Abstract

<P>PROBLEM TO BE SOLVED: To provide a logic-in memory circuit using a magneto-resistance effect element. <P>SOLUTION: As shown in the network 122 of an AND operation of Fig. (b), a TMR network 120 using a TMR element as a magneto-resistance effect element is constructed as a variable resistance element network such that two kinds of variable resistance elements R whose resistance values are decided by an external input X or a storage input Y are connected in series and in parallel, and a total resistance value R<SB>total</SB>is minimum in the case of a specific combination, i.e, a current value I becomes maximum. When the resistance values of the variable resistance elements R decided by the external input X and a storage input Y are R<SB>xi</SB>, R<SB>yi</SB>(i=0, 1, and 2), as shown in FIG. (d), the current value I flowing on the network is decided by the values of x and y. When a threshold value is set between I<SB>0</SB>and I<SB>1</SB>by a threshold value detector 160, an AND operation can be realized. The result of the operation is outputted as a voltage value through a current-voltage converter 170. <P>COPYRIGHT: (C)2005,JPO&NCIPI

Description

  The present invention relates to a logic-in-memory circuit using a magnetoresistive effect element, which is a variable resistance element that can be stored.

  In recent years, performance degradation due to an increase in wiring delay and wiring area is becoming a serious problem on integrated circuits where performance of computing units, memories, and the like is increasing due to rapid progress in microfabrication technology. From the viewpoint that many of these wiring problems are caused by the conventional architecture that separates the arithmetic function and the storage function and configures the circuit, as a method to fundamentally solve the wiring problem, the arithmetic unit and the memory are integrated. Focusing on the logic-in-memory architecture that configures the system, and using the ferroelectric device used as the storage element of the ferroelectric memory (FeRAM), the calculation function and the storage function are integrated in a compact manner The inventors have proposed a high-performance logic-in-memory circuit technique (see Non-Patent Documents 1 and 2).

(Logic in memory circuit)
FIG. 1 shows a normal VLSI 10 (see FIG. 1A) composed of a logic circuit and a memory circuit and a VLSI 20 (see FIG. 1B) of a logic-in memory circuit.
Processing in the normal VLSI 10 is performed by reading data from the memory circuits 11 and 12 via the bus 15, calculating with the logic circuits 17 and 18, and writing the calculation result into the memory circuits 11 and 12 via the bus 15. Is called.
The logic-in-memory VLSI 20 using the non-volatile device shown in FIG. 1B distributes the storage function in the arithmetic circuit 22 and localizes the data transfer between the arithmetic unit 24 and the memory 25, so that the global wiring ( Data transfer bottlenecks on the bus 15) can be greatly reduced. In addition, since the nonvolatile memory is used, stored data can be held without a power supply voltage, so that the leakage current of the storage element can be significantly reduced and a significant reduction in power consumption can be expected.
The logic-in-memory method is useful for configuring a highly parallel computing system that eliminates data transfer bottlenecks because it can significantly reduce the amount of wiring by localizing the data transfer between the computing unit and the memory. Pipeline multipliers and fully parallel associative memories can be realized in a compact manner.
In these logic-in-memory circuit technologies, an arithmetic circuit is configured using a nonvolatile memory device, and therefore, the application range greatly depends on the characteristics of the device. Therefore, in order to expand the application range of the logic-in-memory architecture, it is important to utilize the excellent characteristics of not only ferroelectric capacitors but also various nonvolatile memory devices in the logic-in-memory circuit.

(TMR element)
As element technology of a new logic-in-memory circuit, a magnetoresistive effect element, in particular, a ferromagnetic tunnel magnetoresistive effect (TMR) element which is mainly used as a memory element of a ferromagnetic memory (MRAM) can be used. Possible (see Non-Patent Document 3). A TMR element, also called a ferromagnetic tunnel junction element (MTJ element), utilizes a spin tunnel magnetoresistance effect that changes its resistance value depending on the spin (magnetization) direction of a ferromagnetic material, thereby allowing data access to a magnetic memory element. Since it can be electrically performed, any excellent storage function of magnetic storage that can be nonvolatile, nondestructive reading, high-speed access, and infinite rewriting can be realized on an integrated circuit.
FIGS. 2, 3-1, and 3-2 show the structure and operation of the TMR element. As shown in the sectional view of FIG. 2A, the TRM element 30 includes a ferromagnetic layer (free layer) 32 whose spin (magnetization) direction is changed by an externally applied magnetic field, an ultrathin nonmagnetic layer 34, an externally applied magnetic field. And a ferromagnetic layer (fixed layer) 36 having a constant spin direction, and has a three-layer structure of ferromagnetic layer / nonmagnetic layer / ferromagnetic layer. In the fixed layer 36, a diamagnetic layer 38 is provided to fix the spin direction of the ferromagnetic layer 36. FIG. 2B is a symbol diagram of the TMR element 30. A major feature of the TMR element is that the electrical resistance of the element 30 increases or decreases abruptly when the spin direction of the free layer 32 is changed by an external magnetic field, as shown in the hysteresis characteristic of FIG. For example, when the spin directions of the free layer 32 and the fixed layer 36 are antiparallel to each other, the tunneling current of the nonmagnetic layer becomes small, so the resistance value becomes small. On the contrary, when the spin directions of the free layer 32 and the fixed layer 36 are parallel, the tunnel current of the nonmagnetic layer 34 increases, and the resistance value increases. Therefore, when the storage data is defined as “1” when the spin direction is parallel and “0” when the spin direction is antiparallel, the TMR element 30 can be regarded as a nonvolatile storage element that retains the storage data as a resistance value.

The write operation to the TMR element 30 is performed by causing a current to flow through the wiring adjacent to the TMR element and generating a magnetic field around the wiring. FIG. 3A shows the write operation. As shown in FIG. 3A, currents I _B and I _W are passed through two orthogonal adjacent wirings (bit line 42 and write line 44), and writing is performed by a combined magnetic field H _B + H _W. Figure 3-1 (b), (c) it is, depending on the direction of flow of current _{I B,} shows that the data "0", "1" is written. As described above, data can be selectively written to the cells on the memory array depending on whether or not the current _IW is supplied to the write line 44, and thus is widely used in the MRAM.
An operation for reading stored data from the TMR element 30 is shown in FIG. Storing data read operation, as shown in FIG. 3-2 (a), by applying a voltage to the TMR element 30 from the bit line 42, by detecting the current _{I R} flowing in accordance with the resistance value of the TMR element 30 (See FIGS. 3-2 (b) and (c)). At this time, the current I _R flowing on the bit line 42, sufficiently smaller than the current I _B flowing at the time of writing, storing data is not destroyed. That is, a nondestructive read operation is possible.
When a memory element is realized by a TMR element on an actual circuit, a magnetoresistive (MR) ratio ((R _max −R _min ) / defined by the maximum resistance R _max and the minimum resistance R _min in FIG. Increasing R _min ) is important for high performance, but currently, a TMR element having a large MR ratio of 45 to 55% has been proposed (see Non-Patent Document 3).
The TMR element can electrically access data to the magnetic memory element by utilizing the spin tunnel magnetoresistance effect in which the resistance value changes depending on the spin (magnetization) direction of the ferromagnetic material. One of the excellent storage functions of non-volatile, non-destructive reading, high-speed access, and infinite rewritability can be realized on an integrated circuit.
Note that Non-Patent Document 4 is one in which a TMR element is used in a logic-in-memory circuit.
Examples of magnetoresistive elements other than TMR elements include giant magnetoresistive elements (GMR elements). The major difference between this GMR element and the above-described TMR element is the physical mechanism by which the resistance value changes due to magnetism. In a TMR element, when a voltage is applied perpendicular to a nonmagnetic layer (tunnel layer, mainly AlO ₂ ), the tunnel current changes depending on whether the magnetization directions in the two ferromagnetic layers are parallel or antiparallel. The resistance value changes due to the tunneling magnetoresistive effect. On the other hand, in a GMR element, when a voltage is applied horizontally to a nonmagnetic layer (mainly using Cu), electrons in the nonmagnetic layer depend on whether the magnetization directions in the two ferromagnetic layers are parallel or antiparallel. Since the degree of scattering changes, the resistance value changes (see Non-Patent Document 5).
The basic operation for the GMR element is the same as that of the TMR element, the read operation is performed by detecting the resistance value, and the write operation is performed by changing the direction of magnetization in the ferromagnetic body by applying external magnetization.

T. Hanyu, H. Kimura, M. Kameyama, Y. Fujimori, T. Nakamura and H. Takasu, “Ferroelectric-Based Functional Pass-Gate for Fine-Grain Pipelined VLSI Computation,” in IEEE Int. Solid-State Circuits Conf . (ISSCC) H. Kimura, T. Hanyu, M. Kameyama, Y. Fujimori, T. Nakamura and H. Takasu, “Complementary Ferroelectric-Capacitor Logic for Low-Power Logic-in-Memory VLSI,” in IEEE Int. Solid-State Circuits Conf. (ISSCC) Dig. Tech. Papers, pp.160-161, Feb. 2003. M. Motoyoshi, K. Moriyama, H. Mori, C. Fukumoto, H. Itoh, H. Kano, K. Bessho and H. Narisawa, “High-Performance MRAM Technology with an Improved Magnetic Tunnel Junction Material,” in IEEE Symp. on VLSI Technology Dig. Tech. Papers, pp.212-213, June 2002. Hiroaki Kimura, Takahiro Hanyu, Mitsutaka Kameyama "Configuration of Logic-in-Memory VLSI Using Nonvolatile Devices", IEICE Technical Report TECHNICAL REPORT OF IEICE ICD 2003-5 (2003-04) Betty Prince, “Emerging Memories-Technologies and Trends,” Kluwer Academic Publishers, 2002.

  An object of the present invention is to provide a logic-in-memory circuit using a magnetoresistive effect element.

In order to achieve the above-described object, the present invention connects a magnetoresistive effect element whose resistance value changes according to a stored value and a transistor whose resistance value changes according to an external input value in series and parallel. A magnetoresistive effect network configured to minimize the resistance value of the network by a combination of a specific stored value and an external input value, and a threshold value for detecting a threshold value of a current flowing through the magnetoresistive effect network A logic-in-memory circuit including a detection circuit, wherein the threshold value detection circuit detects a combination of a specific stored value and an external input value.
The magnetoresistive effect network uses the fact that the magnetoresistive effect element has a smaller change in the resistance value than the transistor, thereby reducing the number of combinations of a specific stored value and an external input value that minimize the resistance value of the network. It can be realized by a magnetoresistive element and a transistor.

The magnetoresistive effect network includes two complementary outputs, and configures a differential pair circuit having the two outputs as current paths, and the threshold detection is performed on each current path in the differential pair circuit. Preferably by comparing the currents flowing through
The differential pair circuit is driven by a clock signal, and compares the current flowing in the magnetoresistive effect network using the accumulated charge with a precharge that accumulates charges in a capacitor connected to a current path. It can also be a dynamic type with an evaluation phase for accumulating charge.
The magnetoresistive effect element constituting the magnetoresistive effect network may be a TMR element.
The magnetoresistive effect element constituting the magnetoresistive effect network can be a multi-value TMR element. At this time, a multi-value input is divided for each weight to form a TMR network, and each magnetoresistive effect network is combined with an input having a weight of 0 to form a differential pair circuit. It is desirable to add the outputs.

  Focusing on the fact that the magnetoresistive effect element can be regarded as a variable resistance element whose resistance value changes depending on the stored data, only in the case of a certain combination of external data and stored data, in combination with a transistor controlled by external input data Since the magnetoresistive effect network having the minimum value is configured and a specific combination is detected and output, the magnetoresistive effect element operates not only as a memory element but also as an arithmetic element in the magnetoresistive effect network. Therefore, the excellent memory function of the magnetoresistive effect element can be integrated into the arithmetic circuit in a compact manner.

Embodiments of the present invention will be described below with reference to the drawings.
FIG. 4 is a diagram showing an overall configuration of a logic-in-memory circuit using a variable resistance element using a TMR element as a magnetoresistive effect element.
The logic-in-memory circuit 100 shown in FIG. 4 has a binary external input X and a storage input Y, stores an arithmetic function for performing an operation f (X, Y) using the inputs X and Y, and a storage input Y Memory function. The calculation result is obtained as a binary output Z. Such a logic-in-memory circuit 100 includes a TMR network 120 including a TMR element having an arithmetic function and a storage function, a threshold detector 160, and a current-voltage converter 170.

For example, as shown in the AND operation network 122 of FIG. 5B, the TMR network 120 is a series-parallel connection of two types of variable resistance elements R whose resistance values are determined by the external input X or the storage input Y, and f (X , Y) = 1, the variable resistance element network is configured such that the total resistance value _Rtotal is minimum, that is, the current value I is maximum. Here, as shown in FIG. 5C, the variable resistance elements constituting the network indicate high resistance when the input data is “0” and low resistance when the input data is “1”. On an actual circuit, a variable resistance element whose resistance value is determined by an external input X is realized by a transistor (MOS transistor), and a variable resistance element whose resistance value is determined by a memory input Y is realized by using a TMR element. For example, when the AND operation circuit 122 shown in FIG. 5B is configured, the variable resistance element R is connected in series and parallel. At this time, if the resistance values of the variable resistive element R determined by the external input X and the storage input Y are R _Xi and R _Yi (i = 0, 1, 2), respectively, as shown in FIG. The value I of current flowing through the network is determined by the value of y. At this time, I becomes the minimum (I ₀ ) when (X, Y) = (1, 1). Here, when the threshold value detector 160 determines the threshold value between I ₀ and I ₁ , an AND operation can be realized. The calculation result is output as a voltage value through the current-voltage converter 170.

まず、図６−２（ａ）のネットワークにおいて通常の例えばカルノー図を用いた簡単化により、図６−２（ｂ）のように変形できる。

さらに、ＴＭＲ素子の抵抗値変化（６０ｋΩ〜９０ｋΩ）はトランジスタ（２０ｋΩ〜１ＭΩ）に比べて小さいため、トランジスタのスイッチング状態によって、ＴＭＲ素子の抵抗値に関係なく大電流が流れる場合は、ＴＭＲ素子を付け足しても大電流の流れる電流パスは変わらない（図６−３（ｂ）参照）。これはＴＭＲネットワーク独自の特徴であり、ＴＭＲ素子とトランジスタの抵抗値の違いを活用することで、論理式一定で回路を簡単化することが可能である。最終的には図６−３（ｃ）（簡単形）のように簡単化できる。 (Simplification of TMR network)
As described above, as shown in FIG. 6A, the external input x and the storage input y shown in FIG. 6A are actually connected in series with a MOS transistor Tr and a TMR element _RT . Use the one connected to. By using the characteristics of this configuration, the TMR network can be made a simpler configuration. This will be described with reference to FIGS. 6-2 and 6-3.
Consider simplifying the TMR network portion from FIG. 6-2 (a) (standard form). In FIGS. 6-2 and 6-3, A and C correspond to external inputs, and B corresponds to storage inputs. FIG. 6-2 (a) shows the following logical expression for calculating the carry of the full adder configured as a TMR network in a standard form.

First, in the network of FIG. 6-2 (a), it can be modified as shown in FIG. 6-2 (b) by simplification using a normal Carnot diagram, for example.

Further, since the change in resistance value (60 kΩ to 90 kΩ) of the TMR element is smaller than that of the transistor (20 kΩ to 1 MΩ), if a large current flows regardless of the resistance value of the TMR element depending on the switching state of the transistor, Even if it is added, the current path through which a large current flows does not change (see FIG. 6-3 (b)). This is a unique characteristic of the TMR network, and the circuit can be simplified with a constant logical expression by utilizing the difference in resistance value between the TMR element and the transistor. Finally, it can be simplified as shown in FIG. 6-3 (c) (simplified form).

(Source coupled logic circuit)
Now, in the logic-in-memory circuit 100 using the variable resistance element network 120 shown in FIG. 7A, a minute resistance value change in the variable resistance element network 120 is detected at a high speed, and the threshold detector 160 is used. It is important to realize the current-voltage converter 170 in a compact manner. Therefore, the inventors focused on the threshold calculation function and high driving capability of the source coupled logic.
As shown in FIG. 7B, a source-coupled logic circuit (differential pair circuit) is configured by MOS transistors Mn and Mn ′ using variable resistance element networks 220 and 220 ′ that perform complementary operations. 7A is a compact implementation of the logic-in-memory circuit shown in FIG. In the circuit shown in FIG. 7B, the threshold operation between the currents I and I ′ is performed by a differential pair circuit composed of MOS transistors Mn and Mn ′, and the respective currents are differential between the transistors Mn and Mn ′. Since it is converted to a voltage value via the pair circuit, current / voltage conversion is performed simultaneously. Thereby, the high-performance logic-in-memory circuit 200 is realized in a compact manner.

In the source coupled logic (TMR source coupled logic) 200 using the TMR network, as shown in FIG. 8, a variable resistance element whose resistance value is changed by an external input or a memory input is connected in series and parallel. Realize complementary arbitrary logic operations. When a complementary two-variable logical operation for performing a logical operation f (x, y) between the external input x and the storage input y is realized, as shown in FIG. 8, x (bar x) or y (bar The variable resistance elements depending on y) are connected in series and connected in parallel so that a large current flows only in one current path according to the combination of x and y shown on the current path. To do. Next, only when x and y are in a specific combination, the upper terminal is connected to z or bar z so that the total resistance value R _total of the variable resistance element is minimum, that is, the current value is maximum, and complementary. The variable resistance element network is configured. Here, the TMR source coupled logic circuit 200 can be configured by realizing the variable resistance elements depending on the external input and the storage input with MOS transistors and TMR elements.

に対応して抵抗値の変化する相補ＴＭＲネットワーク２２０，２２０’を用いている。相補ＴＭＲネットワーク２２０，２２０’を流れる電流Ｉ，Ｉ’は、外部入力および記憶入力によって図９（ｂ）のように変化し、電流Ｉ，Ｉ’間のしきい演算をトランジスタＭｎ，Ｍｎ’で構成する差動対回路の比較で行うことにより、ＡＮＤ演算を実現する。この回路では、ＴＭＲ素子を記憶素子のみならず演算回路の一部として用いるため、記憶と演算を一体化したコンパクトなロジックインメモリ回路２００が構成できる。 FIG. 9 shows an example of an AND circuit using a TMR source coupled logic circuit. In FIG. 9A, complementary AND operations are performed on the transistors Mn and Mn ′.

Complementary TMR networks 220 and 220 ′ whose resistance values change correspondingly are used. The currents I and I ′ flowing through the complementary TMR networks 220 and 220 ′ change as shown in FIG. 9B by the external input and the storage input, and the threshold calculation between the currents I and I ′ is performed by the transistors Mn and Mn ′. An AND operation is realized by comparing the differential pair circuits that are formed. In this circuit, since the TMR element is used as a part of the arithmetic circuit as well as the memory element, a compact logic-in-memory circuit 200 in which memory and arithmetic are integrated can be configured.

[Example]
FIG. 10 shows a complementary TMR network of full adders that can be used in the TMR source coupled logic circuit 200 shown in FIG. The complementary TMR network shown in FIG. 10 has the simplest configuration by a method for simplifying the above-described TMR network. FIG. 10A shows a configuration of a complementary TMR network 230 that calculates Carry (carry), and FIG. 10B shows a configuration of a complementary TMR network 240 that calculates Sum (sum). Yes.
Note that FIG. 10C shows a configuration of a normal logic circuit.

(Dynamic TMR source coupled logic circuit)
The realization by the TMR source coupled logic circuit of FIG. 7A has a problem that power consumption increases because a steady current flows through the circuit. In order to solve this problem, a dynamic TMR source coupled logic circuit 300 shown in FIG. 11 is configured using a dynamic current mode circuit (DyCML).
The dynamic TMR source coupled logic circuit 300 is sandwiched between MOS transistors Mn ₃₁ , Mn ₃₁ ′ and Mn ₃₂ , Mn ₃₂ ′ and Mn ₃₃ , Mn ₃₄ which are turned on / off by clock signals CLK and CLK. TMR network 330, and a capacitor _{C 1, C 2, C L} inserted into the current path.
Circuit 300 of FIG. 11, the clock signal CLK, and accumulating charges in the capacitor _C 1, _{C 2} and "pre-charge", to move the charge from the capacitor _C 1, _{C 2} in the capacitor _{C L,} flowing TMR network 330 There are two phases of “evaluation” in which a threshold operation between the currents I and I ′ is performed by a differential pair circuit. In each phase, operation is performed only by charging / discharging of the capacitor, so there is no steady current and power consumption can be suppressed.
The configuration described above may use other magnetoresistive elements such as giant magnetoresistive elements (GMR elements).

(Multi-value TMR network)
Although the binary TMR network has been described above, a multi-value TMR network can also be configured.
In the multi-value TMR network circuit shown in FIG. 12, the logical operation is divided into three depending on which value of the input is 1, 2, 3 with respect to the 4-value (0, 1, 2, 3) input, Each operation is performed by the TMR source coupled logic circuits 420, 430, and 440, and finally, the adder 460 linearly adds them. The outputs of the TMR source coupled logic circuits 420, 430, and 440 have a weight of 1, 2, or 3, and a final output can be obtained by adding them.
Further, for the inside of the TMR source coupled logic circuits 420, 430, and 440, a multi-value ferromagnetic tunnel junction (MTJ) element is utilized.
A four-value MTJ element 500 is shown in FIG. In the quaternary MTJ element 500, two ferromagnetic layers (free layers (1) and (2)) 510 and 530 in the upper part and the central part are present with the nonmagnetic layer (1) 520 interposed therebetween. The fixed layer 550 and the diamagnetic layer 560 provide magnetism in a certain direction with the nonmagnetic layer (2) 540 interposed therebetween. When this quaternary MTJ element is represented by the same symbol as in FIG. 2B, it is represented as shown in FIG.
The quaternary MTJ element 500 has four resistances as shown in FIG. 13C by changing the spin directions of the ferromagnetic layers (free layers (1) and (2)) 510 and 530 from the outside. Can have a value.
By using this element, a compact circuit can be expected compared with a binary TMR element. For example, T. Uemura and M. Yamamoto, “Proposal of Four-Valued MRAM Based on MTJ / RTJ Structure,” in Proc. IEEE Int. Symp. Multiple- See Valued Logic, pp.273-278, May 2003, etc.

(Application examples)
As an application example of the TMR source coupled logic circuit described above, a CRAM (Computational RAM) can be considered. The CRAM is useful for the configuration of a highly parallel computing system such as image processing. As shown in FIG. 14A, the local memory 612, 622, 632 and the computing units (PE) 614, 624, 634 are connected to each other. Data is read from the memories 612, 622, and 632, and arithmetic operations are performed by the arithmetic units 614, 624, and 634. Regarding CRAM, for example, DG Elliott, M. Stumm, WM Snelgrove, C. Cojocaru, R. Mckenzie, “Computational RAM: Implementing Processors in Memory,” in IEEE Design and Test of Computers, pp. 32-41, Jan. -Please refer to Mar. 1999.
When the CRAM is configured by the TMR source coupled logic circuit, the PEs 712, 722, and 732 and the memory can be integrally configured in the TMR networks 710, 720, and 730, as shown in FIG. As described above, when the TMR source coupled logic circuit is configured, the high parallel operation function and the nonvolatile memory function can be integrated in a compact manner, so that the chip area can be greatly reduced and the nonvolatile memory function of the TMR element can be reduced. Since memory retention is possible without power supply, a significant reduction in leakage current during standby can be expected.

2A is a diagram illustrating a normal VLSI configured with a logic circuit and a memory circuit, and FIG. 2B is a diagram illustrating a VLS of a logic-in-memory circuit. It is a figure which shows the structure and characteristic of a TMR element. It is a figure which shows the operation | movement of (a) writing and (b) reading of a TMR element. It is a figure which shows the input / output structure of a TMR network. It is a figure explaining operation | movement of a TMR network. It is a figure which shows the structure of a TMR network. It is a figure which shows simplification of a TMR network. FIG. 3 shows further simplification of a TMR network. It is a figure which shows the structure of a source coupled logic circuit. It is a figure which shows the structure of a complementary TMR network. It is a figure which shows the structural example of the source coupled logic circuit of AND operation. (A) (b) It is a figure which shows the TMR network structural example of the full addition used for a source coupled logic circuit. (C) It is a block diagram of the conventional full adder. It is a figure which shows the structural example of a dynamic type source coupled logic circuit. It is a figure which shows the structural example of a multi-value TMR network. It is a figure which shows the structural example of a 4-value TMR element. It is a figure which shows the CRAM structural example as an application.

Claims (7)

A magnetoresistive effect element whose resistance value changes according to the stored value and a transistor whose resistance value changes according to the value of the external input are connected in series and parallel to form a network, and a combination of a specific stored value and the value of the external input And a magnetoresistive network in which the resistance value of the network is minimized,
A threshold value detection circuit for detecting a threshold value of a current flowing through the magnetoresistive effect network, and the threshold value detection circuit detects a combination of a specific stored value and a value of an external input. Logic in memory circuit. The logic-in-memory circuit according to claim 1,
The magnetoresistive effect network uses a magnetoresistive effect element having a smaller change in resistance value than a transistor, and reduces the combination of a specific stored value and an external input value that minimizes the resistance value of the network. A logic-in-memory circuit realized by a resistance effect element and a transistor. The logic-in-memory circuit according to claim 1 or 2,
The magnetoresistive effect network has two complementary outputs;
A differential pair circuit having the two outputs as current paths is configured,
A logic-in-memory circuit, wherein the threshold value detection is performed by comparing currents flowing through respective current paths in the differential pair circuit. The logic-in-memory circuit according to claim 3,
The differential pair circuit is driven by a clock signal, and compares the current flowing in the magnetoresistive effect network using the accumulated charge with a precharge that accumulates charges in a capacitor connected to a current path. A logic-in-memory circuit, characterized by a dynamic type having an evaluation phase for accumulating electric charges. The logic-in-memory circuit according to any one of claims 1 to 4,
A logic-in-memory circuit, wherein the magnetoresistive effect element constituting the magnetoresistive effect network is a TMR element. The logic-in-memory circuit according to claim 1,
2. The logic-in-memory circuit according to claim 1, wherein the magnetoresistive effect element constituting the magnetoresistive effect network is a multi-value TMR element. The logic-in-memory circuit according to claim 6,
In the magnetoresistive effect network, a multi-value input is divided into weights other than weight 0 to form a magnetoresistive effect network, and each magnetoresistive effect network is combined with a weight 0 input to form a differential pair circuit. A logic-in-memory circuit configured to output by adding the outputs of each differential pair circuit.

Images